大视场视频全景图生成研究

针对采用传统方法生成的视频全景图存在拼接裂缝和“鬼影”现象的问题，提出了一种新的大视场视频全景图生成算法．该算法基于快速傅立叶变换的图像配准算法，求取两相邻视频帧之间的配准系数，采用改进的线性系统法对配准系数进行全局优化．对存在运动物体的视频图像，采用启发式最优路径搜索算法寻找到图像间拼接的最佳缝合线，并使用多分辨率原理进行图像拼接后的平滑处理.实验结果表明，该算法优化了图像间的配准系数，平滑处理了拼接后的图像，减少了人工性缺陷，提高了视频全景图的真实感.

Research on large field view video panorama

Considering the residual ghosting or blurring problems in video panorama synthesized by traditional approach, a new generation algorithm of large field view video panorama was proposed. Based on image registration algorithm with fast Fourier transform, the parameters of two neighboring frames image registration were calculated, and were globally optimized by advanced linear system method. For video images with moving objects, an optimal stitching line was found by heuristic optimal path search algorithm in mosaics of images, and the images were smoothed by multi-resolution theory after mosaics of images. Experimental results show that the algorithm can globally optimize the image registration parameters, smooth the image with moving objects, reduce artificial disfigurement, and enhance photo realistic aspect of video panorama.